Research · Hugging Face Blog
Meta 发布 CyberSecEval 2：面向大语言模型的网络安全风险综合评测框架 ↗
Meta 发布 CyberSecEval 2，一个开源的 LLM 网络安全风险评测框架，包含五类测试：不安全编码实践生成、提示注入攻击易感性、协助网络攻击请求的合规性、代码解释器滥用倾向、自动化进攻性网络安全能力。所有代码开源，支持社区运行并提交结果至排行榜。
TIAOTECH / NEWS
Stay curious. Stay in the loop.
Selected industry news from AIHOT. Follow advances in AI, products and technology.
AIHOT · Selected summaries with original sources · Nov 4, 2026, 9:00 AM
AI products · Hugging Face Blog
Hugging Face Inference Endpoints Support AWS Inferentia2 Instances ↗
Hugging Face announced the addition of support for AWS Inferentia2 instances in its Inference Endpoints service, covering over 100,000 public models.
AI products · Hugging Face Blog
Hugging Face Spaces Launches Dev Mode ↗
Hugging Face Spaces has introduced the Dev Mode feature, which supports one-click connection to VS Code or SSH for directly editing Space code without requiring git pushes or container rebuilds; this feature is currently in beta and available only to PRO subscribers.
Industry · Hugging Face Blog
Hugging Face and Microsoft Deepen Partnership: Azure AI Studio Adds One-Click Deployment for Open-Source Large Models like Llama 3 and Mistral 7B ↗
Hugging Face and Microsoft announced a deepened strategic partnership at the Microsoft Build conference, adding Llama 3 to the Hugging Face Collection in the Azure Model Catalog.
AI products · Hugging Face Blog
Hugging Face and Dell Jointly Launch Dell Enterprise Hub ↗
Hugging Face and Dell jointly launched the Dell Enterprise Hub, offering a one-stop experience for locally training and deploying open-source large language models (such as Llama 3, Mixtral, and Gemma) on Dell hardware.
AI products · Hugging Face Blog
Hugging Face Announces Full-Stack Integration of AMD Instinct MI300 GPUs ↗
Hugging Face announced its collaboration with AMD to complete full-stack integration of the AMD Instinct MI300 series GPUs (including MI300X) on the Hugging Face platform.
AI models · DeepSeek 官方更新
DeepSeek-V2-0517 Released: Upgraded Instruction Following and JSON Output Capabilities ↗
The deepseek-chat model has been upgraded to DeepSeek-V2-0517. The IFEval Prompt-Level accuracy increased from 63.9% to 77.6%. Instruction-following capabilities in the system prompt area via API have been optimized, enhancing immersive translation and RAG experiences. The JSON parsing rate increased from 78% to 85%, reaching 97% after introducing regular expressions.
Practical tips · Hugging Face Blog
Hugging Face Releases KV Cache Quantization Feature to Support Longer Text Generation ↗
Hugging Face has integrated KV Cache quantization functionality into 🤗 Transformers, supporting low-precision formats such as int2/int4, which significantly reduces memory usage for long-context generation.
Industry · OpenAI News
Ilya Sutskever Leaves OpenAI, Jakub Pachocki Appointed Chief Scientist ↗
Ilya Sutskever will leave OpenAI, and Jakub Pachocki has been announced as the new Chief Scientist.
Research · Hugging Face Blog
Hugging Face Releases Open Arabic LLM Leaderboard (OALL) ↗
Hugging Face and Technology Innovation Institute jointly released the Open Arabic LLM Leaderboard (OALL), integrating AlGhafa.
AI products · Hugging Face Blog
Hugging Face and LangChain Jointly Release langchain-huggingface Toolkit ↗
Hugging Face and LangChain jointly released the partner package langchain-huggingface, providing HuggingFacePipeline.
AI models · Hugging Face Blog
Google Releases Open-Source Multimodal Model PaliGemma ↗
Google has released PaliGemma, an open-source vision-language model family consisting of a SigLIP-So400m image encoder and a Gemma-2B text decoder, which is now available on the Hugging Face Hub.
AI models · OpenAI News
OpenAI Releases GPT-4o Multimodal Flagship Model ↗
OpenAI announced the launch of GPT-4 Omni (GPT-4o), a new flagship model capable of real-time reasoning across audio, vision, and text.
AI models · OpenAI News
OpenAI Releases GPT-4o and Opens It to Free ChatGPT Users ↗
OpenAI has released its latest flagship model, GPT-4o, and is making more features available to free ChatGPT users.
AI models · OpenAI News
OpenAI Releases GPT-4o and Expands Capabilities for Free Users ↗
OpenAI has released GPT-4o and made more features available to free users in ChatGPT.
AI products · Hugging Face Blog
Hugging Face Releases Transformers Agents 2.0 and Splits It into a Standalone Library, smolagents ↗
Hugging Face officially released Transformers Agents 2.0, introducing two new types of iterable agents: ReactCodeAgent and ReactJsonAgent.
Research · Hugging Face Blog
Hugging Face Releases Open Leaderboard for Hebrew Large Language Models ↗
Hugging Face has released an open leaderboard dedicated to Hebrew large language models, featuring four tasks: Hebrew question answering, sentiment analysis, Winograd coreference resolution, and English-Hebrew translation, all using few-shot prompting formats. The evaluations are based on datasets constructed from HeQ, Hebrew Sentiment, a translated version of Hebrew Winograd, and NeuLabs-TedTalks corpora, executed automatically via Inference Endpoints and lighteval.
AI models · Hugging Face Blog
Hugging Face Launches Artificial Analysis LLM Performance Leaderboard ↗
Hugging Face has launched the LLM performance leaderboard published by Artificial Analysis, covering over 100 server-side LLM APIs and comprehensively evaluating quality (MMLU/MT-Bench/HumanEval/Chatbot Arena).
Industry · OpenAI News
OpenAI Bans Accounts Linked to Russian 'Doppelganger' Operation ↗
OpenAI has banned accounts associated with the Russian 'Doppelganger' operation, which uses AI to generate multilingual social media comments, translations, and website copy targeting Ukraine.
Practical tips · Hugging Face Blog
Hugging Face Inference Endpoints Release Custom Inference Processor Supporting Speech Recognition, Speaker Diarization, and Speculative Decoding ↗
Hugging Face has launched a modular custom inference processor on its Inference Endpoints that supports ASR (Whisper), speaker diarization (Pyannote), and speculative decoding (distil-whisper assistant model).
Industry · OpenAI News
OpenAI Bans Accounts Linked to 'Spamouflage' Operation ↗
OpenAI banned accounts associated with the China-linked 'Spamouflage' operation, which used AI to research social media activity, generate posts, and debug a previously undisclosed website.
Research · Hugging Face Blog
Hugging Face and Dottxt Collaborate on Research into Structured Generation to Improve Prompt Consistency ↗
The Hugging Face Leaderboards and Evals team collaborated with Dottxt to discover that LLM evaluation results are extremely sensitive to prompt format, with minor changes causing score fluctuations of more than 10 points for the same model on MMLU.
Research · Hugging Face Blog
StarCoder2-15B-Instruct-v0.1 Released: The First Fully Transparent and Freely Usable Self-Aligned Code Large Language Model ↗
Hugging Face released StarCoder2-15B-Instruct-v0.1, the first fully self-aligned open-source code large language model. It was fine-tuned using 50k instruction-response pairs generated by StarCoder2-15B itself, achieving a HumanEval score of 72.6, surpassing CodeLlama-70B-Instruct. The entire data generation and training pipeline is open-sourced, supporting free commercial use and redistribution.
Technology products · PyTorch Releases
PyTorch 2.3 Release: Support for User-Defined Triton Kernels and Native Tensor Parallel Training ↗
PyTorch 2.3 has been officially released. Core updates include support for user-defined Triton kernels in torch.compile (avoiding performance degradation and graph breaks), and native Tensor Parallelism functionality in PyTorch Distributed, which has been verified for use in training large language models with 100B parameters.